You said "The timescale of mtDNA divergence is already out of whack with the rest of the genome." What's the time scale for the rest of the genome? It seems to me it should be expected to be at least twice as much as that for mtDNA since at least half the instances of mtDNA - those in males - dead end each generation. With perfect mixing and replacement, 50% of the mtDNA instances pass from one generation to the next, while 75% of the autosomal instances do. Imperfect mixing and replacement would make both numbers lower, but the mtDNA number would still remain much lower than the autosomal number, so the coalescence time should still be expected to be much lower.
Thanks for noticing that, it’s leading to something but I haven’t yet described the problem. My apologies for being less than clear.
What you’re describing (you probably already know) is commonly described as the “four-times rule” – the uniparental inheritance and single copy number give mtDNA one fourth the effective size, on expectation, as an autosomal locus.
That’s in a constant-sized population. Which of course we haven’t been. For around the past 100,000 years, African populations were big enough that genetic drift didn’t decrease their genetic diversity markedly. The mtDNA coalesces around 100,000 years before that, compared to more than 700,000 years for the typical autosomal locus – it’s 7 times instead of four. That discrepancy is probably not significant given the huge intrinsic variance of the coalescent. But I don’t think it’s been seriously investigated.
The real problem is that the out-of-Africa timescale for mtDNA is now very short – less than 65,000 years – while the nuclear timescale looks long – maybe up to 140,000 years. Maybe these can also be reconciled; it’s not yet clear. But it’s a problem.